Explainable electronic medical record-based machine learning to predict 1-year incident Protein-Energy wasting in non-dialysis chronic kidney disease: A single-center development study.

BACKGROUND

Protein-energy wasting (PEW) is common in chronic kidney disease (CKD) and is linked to poor outcomes. Early risk stratification may enable timely nutritional intervention.

OBJECTIVE

To develop an electronic medical record (EMR)-based machine-learning model for 1-year incident CKD-PEW and report transparent, model-agnostic performance.

METHODS

We retrospectively analyzed 500 adults with CKD managed between July 2019 and May 2021. Patients with PEW (defined by ISRNM criteria) at baseline were excluded. Thirty-one routine clinical/laboratory variables were used to train and evaluate five algorithms-logistic regression (LR), support vector machine (SVM), decision tree, random forest (RF), and LightGBM-under a random 75/25 split (382 training, 118 validation). Model performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC) with 95% confidence intervals (CI), Brier score, calibration plots, and decision curve analysis (DCA). Shapley additive explanations (SHAP) were used for interpretability.

RESULTS

On the hold-out validation set, the optimized Light Gradient Boosting Machine (LightGBM) achieved the highest discrimination with an AUC of 0.95 (95% CI: 0.91-0.99). The model also demonstrated excellent probability calibration with a Brier score of 0.095.

CONCLUSIONS

Using routine EMR data, we developed an explainable model that demonstrates high discrimination for predicting 1-year incident CKD-PEW on a hold-out validation set. The workflow provides a pragmatic foundation for subsequent independent validation, calibration assessment, and decision-curve analysis before clinical use.

CLINICAL TRIAL NUMBER: ChiCTR2100046679.